How to fix cvxpy output of arrays?

Question

I am trying to write a portfolio optimization problem using cvxpy.

initial_weights = [0.045, 0.035, 0.024, 0.028...]
rets = (np.log(data/data.shift(1))
w = cvx.Variable(38)
ret = np.sum(rets.mean()*w)*252
prob = cvx.Problem(cvx.Maximize(ret), [cvx.sum_entries(w)==1, w>0.02, w<0.06])

result = prob.solve() After solving, w.value is of the form [[0.02], [0.02], [0.04]...]

I need to add more constraints like abs(list_final_weights-w)>0.005, but that will be possible if w.value is of the form of array of values instead of array of arrays. How can I fix this error?

What's `x` in the constraint? – Adarsh Chavakula May 31 '19 at 13:54 — Adarsh Chavakula, May 31 '19 at 13:54
That was an error. x is the variable w. – Ray234 May 31 '19 at 14:03 — Ray234, May 31 '19 at 14:03

Adarsh Chavakula · Answer 1 · 2019-05-31T14:23:00.743

0

cvxpy does element-wise operations for subtraction, as long as they're of the same shape. You could convert your flat array into the appropriate shape that the solver needs - In this case it's (N, 1) instead of a flat array of shape (N,).

list_final_weights = np.reshape(list_final_weights, [len(list_final_weights), 1])

Then you should be able to do the subtraction for the constraint:

list_final_weights - w >= 0.005

Also cvxpy does not allow hard > or < inequalities. Only <= and >=. https://www.cvxpy.org/tutorial/intro/index.html#constraints

edited May 31 '19 at 14:23

answered May 31 '19 at 14:14

Adarsh Chavakula

1,509
19
28

I need the absolute difference to be >=0.005 but when I do abs(list_final_weights-w)>0.005, I get a bad operand error. Is there any other way to write this? – Ray234 May 31 '19 at 14:33
use `np.abs` instead of the inbuilt `abs`. The inbuilt one cannot handle arrays. – Adarsh Chavakula May 31 '19 at 14:40
np.abs gives the same type error: bad operand type for Add Expression – Ray234 May 31 '19 at 14:43
Can you share the line of code that raised the error? – Adarsh Chavakula May 31 '19 at 14:45
const = [cvx.sum_entries(w)==1, w>=0.02, w<=0.06, np.abs(w-list_final_weights)>=0.005] – Ray234 May 31 '19 at 14:53
The issue is not with `np.abs`. It's with the first constraint. Try doing `cvx.sum(w)==1` or `np.sum(w)==1` and see if you get the same error. – Adarsh Chavakula May 31 '19 at 15:10
Still the same error. TypeError: Bad operand type for abs(): 'AddExpression' – Ray234 May 31 '19 at 15:22
Last ditch effort - use `cvxopt.abs` instead of `np.abs`. Even if that doesn't work, I am not sure what else can be done here except contacting the developers of the library. – Adarsh Chavakula May 31 '19 at 15:31

How to fix cvxpy output of arrays?

1 Answers1